摘要
针对工件缺陷种类多样和特征不明显,造成机器视觉识别精度不稳定的问题,提出了一种基于深度学习网络模型与大数据自动训练的工件缺陷识别算法,并以软件工程来实现其功能。首先,对存在缺陷的工件进行取样,采集缺陷图像,建立识别标准。然后,基于深度神经网络模型和缺陷标准图像,进行模型训练,采取分批训练,逐步迭代收敛的方式,达到准确识别工件缺陷目的。最后,基于Python语言与TensorFlow框架实现深度神经网络模型,并将模型移植到C++平台调用,嵌入到商业版本软件中,实现算法的落地应用。实验测试结果显示,相对于已有的缺陷识别技术而言,本文算法具有更高的识别准确性,可为机器视觉软硬件设备提供技术基础。
In order to solve the problem that the recognition accuracy of machine vision is unstable due to the variety and characteristics of workpiece defects, a workpiece defect recognition system based on deep learning network model and large data automatic training is proposed in this paper, which is implemented in the form of software engineering. Firstly, the defective workpiece is sampled, the defective image is collected, and the recognition standard is established. Then, based on the depth neural network model and defect standard image, model training is carried out, batch training is adopted, and iterative convergence is adopted step by step to accurately identify the defect of the workpiece. Finally, the deep neural network model is implemented based on Python language and TensorFlow framework, and the model is transplanted to C++ platform call, embedded in commercial version software, to realize the landing application of the algorithm. The experimental results show that the system is conducive to the stable identification of workpiece defects, and lays the foundation of algorithm and software for machine vision hardware and software equipment.
作者
黄寅
HUANG Yin(Chuzhou Branch School,Anhui Open University,Anhui Chuzhou 239000,China)
出处
《齐齐哈尔大学学报（自然科学版）》
2020年第2期21-24,39,共5页
Journal of Qiqihar University(Natural Science Edition)
